DeepSeek Drove Its Competitors Into an Alliance: OpenAI, Google, and Anthropic Join Forces

OpenAI, Google, and Anthropic have begun collaborating to identify cases of unauthorized use of their models. The companies, normally fierce rivals, are now sharing data with Microsoft also taking part.

This is about so-called distillation — a technique where an existing model trains a new one to replicate its capabilities. The method itself is widely used across the industry to create more compact and efficient versions, and developers often allow customers to use it for non-competitive purposes. But the problem arises when the technology is used to copy proprietary work without permission.

In January 2025, Chinese company DeepSeek released its R1 model, which caused a major stir in the industry. But shortly after, Microsoft and OpenAI began investigating whether the startup had used unauthorized data extraction from U.S. models. By February, OpenAI directly accused DeepSeek of trying to “free-ride on OpenAI’s and other leading U.S. AI labs’ developments.” Around the same time, Anthropic went further and named three Chinese AI labs — DeepSeek, Moonshot, and MiniMax — which, according to the company, had been extracting capabilities from its Claude model via distillation. It claims more than 16 million interactions were created through 24,000 fake accounts.

Most Chinese AI models have an open distribution model, meaning any company or developer can freely download the base system and run it on their own infrastructure. U.S. companies have gone a different route — their models are closed, with paid access intended to help recoup the hundreds of billions of dollars poured into data centers and infrastructure. If Chinese competitors can replicate those capabilities for pennies using distillation of closed models, the entire economics of the American AI business collapses.

The framework is borrowed from cybersecurity, where competing companies have long shared data on attacks and adversary tactics. The approach here is the same: identify cases of distillation, determine who’s responsible, and try to prevent unauthorized access. But the companies aren’t yet fully sure what they can actually share with each other without violating antitrust laws — even though the stated goal is combating the Chinese threat. So far, none of the three companies have provided public evidence of how heavily China’s AI industry actually relies on distillation.

For U.S. companies, this isn’t just a business issue. Anthropic warns that distilled models often lack built-in safety guardrails that prevent AI from being used to create biological weapons or carry out other harmful actions. DeepSeek continues working on a new version of its chatbot, and the industry is closely watching whether distillation will be used again. China itself has not officially responded to the allegations — most local models remain open source, and from their perspective, nothing illegal is happening.

What do you think — is this a real threat to AI development, or just a move by the big players to protect their commercial interests under the guise of fighting the “Chinese threat”? Share your thoughts in the comments.